b200 test

Author: Oseltamivir

.github/workflows/eval-tmpl.yml

@@ -146,6 +146,7 @@ jobs:
           RUN_MODE: eval
           # Optional: structured filename if runner chooses to use it later
           EVAL_RESULT_BASENAME: ${{ env.EXP_NAME }}_${{ env.PRECISION }}_${{ env.FRAMEWORK }}_${{ runner.name }}
+        working-directory: repo
         run: |
           bash ./runners/launch_${RUNNER_NAME%%_*}.sh
 

runners/launch_b200-nvd.sh

@@ -35,9 +35,8 @@ else
   export NUM_PROMPTS=$(( CONC * 10 ))
 fi
 
-cd repo
 ls -lt
-rm -rf /home/nvadmin/actions-runner/_work/InferenceMAX/InferenceMAX/gpt-oss-120b/
+sudo rm -rf /home/nvadmin/actions-runner/_work/InferenceMAX/InferenceMAX/gpt-oss-120b/
 
 docker run --rm --init --network host --name $server_name \
 --runtime nvidia --gpus all --ipc host --privileged --shm-size=16g --ulimit memlock=-1 --ulimit stack=67108864 \